Solidification phase growing during simulation

[Moritz]

Hello everyone!

I want to set up a simulation of the solidification of a steel, which contains high melting oxide particles. As the oxides solidify as the first phase during cooling and solidification, they function as seeds for the further solidification and forming of the initial microstructure.

I am not sure, if MICRESS generally offers the possibility to simulate particles (e.g. Al2O3), which form during cooling and function as solidification seeds and also grow during the cooling.

That is why i am posting this question here? Has somebody maybe experiences with similar questions?

If there is no direct method to do it (let seeds grow withe the microstructure), is there maybe a way to implement a model through several steps? I thought about adding the oxide particles as an initial microstructure manually at differend positions. Is this the only way?

I am looking forward to your replies

Best,
Moritz

[Bernd]

Hi Moritz,

Yes, MICRESS offers the possibility to account for curvature of the substrate during nucleation at interfaces. By specifying distinct phases as substrate and reference phase (Al₂O₃ and LIQUID in your case), the curvature of the substrate (i.e. the growing particle) is defined and automatically added to the nucleation undercooling. From the upcoming version 7.1 there will be an explicit switch for this curvature contribution.

For simulation of e.g. Al₂O₃, aluminium and oxygen must be included as elements, and you need to define some initial positions or nucleation conditions for the seed particles. Then, depending on the oxygen and Al content of the melt and on temperature, particles will grow or shrink, and eventually serve as nucleation sites. The curvature effect on nucleation undercooling will favor nucleation on bigger particles and will be taken into account correctly even if the seed particles are smaller than a numerical grid cell ("small grain").

I personally did similar simulations for nucleation of graphite in grey iron on MnS particles many years ago. In this case, nucleation of MnS and graphite occurred at similar temperature ranges, so that there was a strong effect of small composition variations on this nucleation scenario. For Al₂O₃ in steel this may be different because these nucleants already form at very high temperatures, and thus may be formed already during melt production rather than during solidification.

Of course, you also can specify the seeding particles by putting them at explicit or random positions with given sizes, and let them grow further before solidification starts. However, in case they won't change size significantly during this process, you rather should simplify your approach and use the "seed_density" model instead of explicitly simulating the seed particles (which then would lead essentially to the same results without having to include Al and O into the simulation).

Bernd

[Moritz]

Hi Bernd and to all,

thank you very much for your reply. This was the approach I had in mind.

I tried to implement this by using the example T002_AlCu_Equiaxed and adjusting the following aspects:

- using a Thermocalc-database for the chemical composition I want to simulate (FeCrAlMo steel with oxygen in it)
- adding the components to the model
- adding the expected phases (LIQUID, BCC_A2, CORUNDUM)
- adding a CORUNDUM particle manually in the middle of the model to test the behaviour of the solidification

I can start and run the simulation, but during this time, several things happen, which are not expected.

First of all, I can see this in the .log file:



I am not sure, why there are any other elements in CORUNDUM than aluminium and oxygen.



As I tried to improve the simulation several times, I tried to specify allowed compositions, to forbid any other element in CORUNDUM beside aluminium and oxygen:



Since I don't have a vast amount of data about the all necessary parameters, I sometimes used "dummy" values. My goal was to get the simulation running and then further adjust the parameters to correct or realistic values. I suspect that this approach might lead to problems, for example with diffusion coefficients that are way too fast, but I am not sure.

The result looks like this:


Obivously, this does not look like a microstructure which is to be expected. Additionally to the problems I have with the composition, I think there is a problem which the Thermocalc data regarding the phases. I already read in this forum about, e.g. demixing and this leading to differend errors and warning from Thermocalc and MICRESS. I was not able to transfer the posted solution to my own problem, but I am fairly confidend, that the problem with my simulation is somewhere with Thermocalc and the data used by it.





I attach my .dri file and also .scr, .log and phas files, so it becomes more clear what I mean.

Thank you for any replies!

Best,

Moritz

[Bernd]

Dear Moritz,

There is a subtlety with the phase "CARBORUNDUM" when MICRESS tries to automatically get the composition limits for each phase and define the corresponding elements as "stoichiometric: Normally, one would expect this phase to be modeled with 2 sublattices, where one is completely occupied by Al and one by O, respectively. However, in the database they do not use elements but a species "Al2O3" instead.

For this reason, you get the following warning so that you can check whether MICRESS did everything correctly or not:

Warning: Species in phase LIQUID are not elements!
Please use "limits" to manually set concentration limits!


In phase LIQUID component O is really stoichiometric.
In phase CORUNDUM component MO is really stoichiometric.


Beginning of initialisation
***************************
Grain number 1 set

Start Composition and Limits for quasi-equilibrium
--------------------------------------------------
FE in LIQUID: 90.6550 at% (>0 - 100.000at% )
CR in LIQUID: 1.76900 at% (>0 - 100.000at% )
AL in LIQUID: 1.73482 at% (>0 - 100.000at% )
MO in LIQUID: 0.922959 at% (>0 - 100.000at% )
O in LIQUID: 4.91817 at% (0 - 0at% )
FE in BCC_A2: 32.3288 at% (>0 - 100.000at% )
CR in BCC_A2: 29.8420 at% (>0 - 100.000at% )
AL in BCC_A2: 0.276315 at% (>0 - 100.000at% )
MO in BCC_A2: 34.8157 at% (>0 - 100.000at% )
O in BCC_A2: 2.73723 at% (>0 - 75.0000at% )
FE in CORUNDUM: 16.3823 at% (0 - 0at% )
CR in CORUNDUM: 13.4862 at% (0 - 0at% )
AL in CORUNDUM: 10.3438 at% ( 40.0000 - 40.0000at% )
MO in CORUNDUM: 0.00000 at% (0 - 0at% )
O in CORUNDUM: 59.7877 at% ( 60.0000 - 60.0000at% )

Then, it finds O and MO to be stoichiometric, which is true. But essentially, all elements are stoichiometric, including those who do not exist in this phase (i.e. they have composition 0).

In the following list you see, that the start compositions are not correct (but here you should not worry about that because internally they use the species AL2O3...). What you should worry about in MICRESS are the solubility ranges, i.e. the max and min allowed values, but they are correct as you can see.

Thus, the only thing you should definitively do is define all elements as stoichiometric:

# Concentration solver
# --------------------
# Factor for diffusion time stepping? (0.0 < factor < 1.0)
0.95000
#
# List of phases and components which are stoichiometric:
# phase and component(s) numbers
# List of concentration limits (at%):
# <limits>, phase number and component number
# List of penalty conditions:
# <penalty>, phase 1, phase2, component number
# List for ternary extrapolation (2 elements + main comp.):
# <interaction>, component 1, component 2
# Switches: <stoich_enhanced_{on|off}> <solubility_{on|off}>
# List of relative criteria on phase composition
# <criterion_higher | criterion_lower>, phase No 1, phase No 2, component No
# List of sublattice order conditions:
# <ordered|disordered>, phase , sublattice 1, sublattice 2
# List of source changes for diffusion data
# <switch_diff_data>, Phase-No., reference phase
# Switch: Add composition sets for calculation of diffusion/volume/enthalpy data
# <diff_comp_sets | vol_comp_sets | enth_comp_sets>, phase list
# End with 'no_more_stoichio' or 'no_stoichio'
2 1-4
no_more_stoichio

However, there is another thing which makes your simulation crash: You did not define any diffusion in LIQUID...

Bernd

[Moritz]

Dear Bernd,

thanks again for your helpful and detailed reply.

I defined all elements as stoichiometric as you showed in your reply and also added diffusion in the liquid phase. I must have accidently deleted it somehow.

Unfortunately, the simulation still shows multiple error messages. With the information already available in this forum (for example from this threat viewtopic.php?f=9&t=562&p=2347&hilit=tr ... ases#p2347), I was able to understand a little bit mor eabout the error.



It seems that there is a problem with the interaction between the phases 0 and 1, in my case LIQUID and BCC_A2. Unfortunately, I am not sure, what exactly the problem is.

The new result of the simulation looks like this:



As before, the driving-file and other informations are attached.

Thank you very much for your help!

Best regards,

Moritz

[Bernd]

Hi Moritz,

being short in time, I'll give you a fast answer with hints for avoiding the problems (which may be not complete though...).

1.) Given the high heat extraction rate, you get a cooling rate in the order of -1.E5 K/s. Thus, microstructures are expected to be very fine, while your grid resolution is 1 µm. I guess you will have to go at least down to 0.1 µm, if not further!

2.) In view of the corresponding short time scale, updating intervals for enthalpy, diffusion coefficients, etc. are much too big.

3.) Given that the alloy you are using is not "dilute" anymore, redistribution using the "multi-binary extrapolation" mode can fail ("Demixing..."). You should switch to "diagonal extrapolation" using the keyword "diagonal" in the numerical parameters for concentration section (where you define stoichiometric phases).

4.) The stabilisation value for interface 0/1 (extra input after interface energy) is working only if it is larger that the interface energy. You should use a value which is ~10 times bigger.

Bernd

[Moritz]

Hi Bernd,

thanks for your reply. As you know, I was busy working on other models and problems.

As shown in this post (viewtopic.php?f=9&t=748) my colleagues and I are working on a model for the solidification of an ODS steel. For now, we achieved good results with a simulation that is focused on the columnary and then equiaxed solidification under thermal conditions modeled after the LPBF process.



The current model does not include the full chemical composition of the researched alloy, as oxygen is missing. As we discussed earlier in this thread, simulations with oxygen produce a number of error messages and weird results.

Obviously it is still our goal to model also the aspect of the formation of oxide particles. For this purpose, I took all your suggestions and worked them into our new model. Unfortunately, we still can observe weird results. Is there something else we can do to improve this simulation or to erradicate any mistakes?



As always, I attach the driving file for more details of the simulation.

Best regards and thank you very much

Moritz

[Bernd]

Dear Moritz,

I can see several potentially problematic input details which could explain the problems you have. I will start with the probably worst:

1.) You did not define any diffusion in the LIQUID phase! As solid diffusion is much too slow (in view of the fast cooling), only a massive transformation is possible (but not compatible with the assumption of diffusion control by 'mob_corr'...)

2.) You have chosen interfacial energies for the 0/1 and 0/2 interface which differ by a factor of 100. This is not only completely unrealistic, but also can easily lead to numerical issues at triple junctions. Typically, differences in a triple junction should be smaller than a factor of 2-3.

3.) As already indicated further above in this thread, interface stabilisation only works if the stabilisation value is bigger than the interface energy itself. This is not the case (thus stabilisation does not work currently)

4.) You are using a strange way of defining an initial microstructure (analytical_curvature, voronoi). At the moment, I am not 100% sure how MICRESS behaves with this combination, but I would expect that there should be no LIQUID phase left...

5.) I don't see the point in using gradient boundary conditions for phase_field in bottom and top direction

6.) Depending on the exact cooling conditions which result from reading temperature profiles from file, the current spatial resolution of 0.06 µm may be still too coarse for getting reasonable results. For typical L-PBF conditions (e.g. v=1m/s, P=200W, r=75µm) I usually need Δx=0.01 µm.

7.) With "big" initial seeds already existing at the beginning, you should better use initialisation (smoothening) of the interface ("#Number of steps to adjust profiles of initially sharp interfaces" ~10-30)

8.) I would advise you to include also the solid-solid phase interaction (BCC/CORUNDUM). Otherwise, a "constant-K" approximation is used. This per se is not a problem, but the "local" nature and handling of this type of approximated phase-diagram data may lead to performance losses in case a considerable amount of triple junctions would form.

Best wishes and good luck
Bernd